Official Phenom Reviews Thread

[Epsilon84]

LOL, so AMD expects a 31.4% performance increase from a 13% clockspeed increase?!

The idiots probably benched Yorkfields at those frequencies and used those scores as their own... :p

[KTE]

If you own or have seen someone try a Phenom X4 on a older AM2 platform, you would know they have serious problems everywhere in typical functioning and need a new BIOS which there isn't ATM. GPU performance is horrible along with them compared to an X2 chip in the same board. This is of 23rd Nov rather than 10-16th Nov of the review experimentations. You can't even set RAM timings like SPD without BSODs on the best of ASUS' AM2 board, so those "reviews" which use an AM2 board could've easily had bugged lower numbers in absolute contexts. The new 790 boards should show no problems though apart from Gigabyte's Overdrive support. I have both a new AM2+ and an old AM2 board to check this out very soon.

So, what do the people here think of a common air/water oc for Phenom 9500/9600? Lets hear some educated guesses.

[justapost]

About pricing

Source

Nice 9500 for 200€, i'll order one now.

SSE4a consists of 4 instructions, none of the 4 are related to Intel's SSE4 instruction set. I have not seen any data showing any information on SSE4a compiled software vs non-SSE4a compiled software.... but a resonable assumption is that it does not provide a huge improvement otherwise AMD would have advertised it as a significant selling point.

Maybe a newer version of tmpegenc or divx will have sse4a support. I don't expect much. I'm just curious.

Why is the 9700 and 9900 so much better when the frequences are just a bit higher? (It seems more effective)

Looks like beside they used 7% for a 100MHz clock increase they expect a 10% higher performance with the new stepping.

I have both a new AM2+ and an old AM2 board to check this out very soon.

My board got delayed, but i will order a cpu now. I already downloaded win2008rc0 (32 and 64 bit) for comparison and installed it on a c2d notebook (hp 6820s). My primary goal is to compare the new virtual machine support and compare it to xen but as a sidefeffect i'll get a comparison of os-dependent speed improvements.
Hope i'll find the time to bench all the different configurations i have in mind.
I plan to run the whole thing on am2 and am2+ again, to get an impression how the new virtualisation functions on phenoms work.

[BeardyMan]

Lately there has been a lot of BS(Dave_Graham where are you?) floating around about the magical stepping which will bring enormous performance improvement over the previous "crappy" stepping.
Which one is it? B1A, B2, B3, B4, B6, B-Complex, B92?

And why would AMD need that? Performance is fine, if not good for the price.
Link

[Periander6]

And why would AMD need that? Performance is fine, if not good for the price.
Link

The image you linked that was from here:

http://www.pcper.com/article.php?aid...e=expert&pid=8

The overall 3DMark06 scores show the Phenom is more positive light than they should because of the different graphics cards at work on the two testing setups -- remember the Phenoms were running dual Radeon HD 3850 cards while the Intel Core 2 and AMD Athlon CPUs were running a single 8800 GTX card. The CPU tests aren't affected though and once again the Phenoms split the pack of existing dual- and quad-core CPUs.

Nice job cherrypicking a graph without noting that. Pure BS really.

That review's conclusion:

http://www.pcper.com/article.php?aid...=expert&pid=12

I have no doubts that many readers of this review fill find it disappointing that AMD's Phenom processors were not competitive with Intel's high-end quad-core processors. It's hard to hide my own disappointment as I personally really wanted AMD to do well - competition makes the world go 'round and prices go down; always good things in my book. The Phenom launch isn't a total loss though thanks to the aggressive pricing that AMD is pinning on these initial CPUs; that will appeal to many enthusiasts.

I can't help but draw the analogy of the Phenom launch to that of the Radeon HD 2900 XT: it was a technologically advanced GPU that had scaling issues that kept it from competing at the high end of its market thus forcing AMD to drop prices to stay in the game. Now that AMD's HD 3800 series of GPUs has seemingly fixed that problem, they are presented another with the Phenom desktop processor. Hopefully AMD will make as quick a turnaround on this project as they did on the GPU side and we'll see the Phenoms truly compete at all price ranges.

[Motiv]

And why would AMD need that? Performance is fine, if not good for the price.
Link

indeed your one cherry picked benchmark shows an overclocked phenom winning in 3dmark. Why not post links to every other benchmark from that review?

like this Link

or even the full review Link

[JumpingJack]

And why would AMD need that? Performance is fine, if not good for the price.
Link

Yeah, if you like the slower of the choices.

How about some 3D gaming benches designed to really understand the potency of the CPU as a game code cruncher:
http://www.matbe.com/articles/lire/5...500/page15.php

Synthetics, while interesting ... are not the best way to understand real world performance. Especially 3DMark, which is GPU weighted -- considering they had 2 cards working, is it surprising to see a decent score in 3DMark?

[JumpingJack]

Maybe a newer version of tmpegenc or divx will have sse4a support. I don't expect much. I'm just curious.

The newer version of DivX supports SSE4, they touted that heavily. But SSE4a, I have not seen any public support for these instructions. SSE4a is a different set of instructions, I doubt it will be long before a version comes up with support. I am with you, it would be interesting to see what it does.

[KTE]

The instructions in SSE4 really help one another to remove cache and memory write back bottlenecks so you see a marked difference. AFAI think SSE4a won't make too much if any "performance" difference in the desktop environment. It's missing added instructions which can give the processing delay a boost and those are supposed to come with either one of Shanghai or Deneb. For these, I expect they will only hasten floating point data by removing L1 cache bottlenecks.

You can read a little about the instructions here:
http://forums.amd.com/devblog/blogpo...8051&catid=271
http://forums.amd.com/devblog/blogpo...threadid=88480

[JumpingJack]

The instructions in SSE4 really help one another to remove cache and memory write back bottlenecks so you see a marked difference. AFAI think SSE4a won't make too much if any "performance" difference in the desktop environment. It's missing added instructions which can give the processing delay a boost and those are supposed to come with either one of Shanghai or Deneb. For these, I expect they will only hasten floating point data by removing L1 cache bottlenecks.

You can read a little about the instructions here:
http://forums.amd.com/devblog/blogpo...8051&catid=271
http://forums.amd.com/devblog/blogpo...threadid=88480

Thanks.... I know very little about the details of these instructions.

Jack

[Schneider]

Why is the 9700 and 9900 so much better when the frequences are just a bit higher? (It seems more effective)

Have to remember that the 9500 & 9600 have 1.8Ghz Northbridges while the 9700 has a 2.0Ghz northbridge and the 9900 might just have a 2.2Ghz northbridge. I'm pretty sure the extra speed on the L3 cache there would help with the scaling.

[Donnie27]

Have to remember that the 9500 & 9600 have 1.8Ghz Northbridges while the 9700 has a 2.0Ghz northbridge and the 9900 might just have a 2.2Ghz northbridge. I'm pretty sure the extra speed on the L3 cache there would help with the scaling.

One site complained about only running at 2GHz for the 9700.

[Shintai]

Have to remember that the 9500 & 9600 have 1.8Ghz Northbridges while the 9700 has a 2.0Ghz northbridge and the 9900 might just have a 2.2Ghz northbridge. I'm pretty sure the extra speed on the L3 cache there would help with the scaling.

They still wont scale way beyond 100%...

100% overall scaling in those programs they list. Would require everything to be that faster. GFX, HD, memory, cache speeds, core speed , HT links and so on.

Its simply BS and lies in the shape of marketing.

9500 vs 9600 they say 7%, yet the core speed only goes up 4.5%. And everything else is the same.

[JumpingJack]

Have to remember that the 9500 & 9600 have 1.8Ghz Northbridges while the 9700 has a 2.0Ghz northbridge and the 9900 might just have a 2.2Ghz northbridge. I'm pretty sure the extra speed on the L3 cache there would help with the scaling.

This is the best explanation so far, but I still have a hard time believing that just opening up the BW and lowering latency has that much impact.... however, if it does... then the L3 cache is really doing some damage and will be a critical path in future revisions.

[informal]

JJ,that explanation makes the most sense ,but it would imply(like you said) that L3 is a major bottleneck for K10,which is unlikely.It can make a big difference if NB is ran at the same as core speeds,but we need more tests to verify this.K10 looks and performs quite well,but looking at AMD section and some problems people have with early boards/BIOSes,it seems as the whole platform is still a diamond in a rough(problems with changing basic stuff in BIOS for instance).

[Hornet331]

nice review, it shows that that amd tricore can only competent with intels midrange segment... (trades blows with the E6750)

[informal]

Primus_Sucks,welcome to the XS and thanks for the links.

[Epsilon84]

Excellent link Primus_Sucks!

Hmm, so Phenom X3 @ 2.3GHz ~= E6750/E6850 in multithreaded performance.

Hopefully AMD can get clockspeeds much higher with B3 otherwise Phenom X3 will be stuck battling it out with the higher clocked Intel C2Ds.

[BeardyMan]

according to this post, by end user review Phenom is clock per clock good afteral

in supcom 2.64ghz + ~ 3GHz quad intel

that's not bad

LINK

also look at the 3dmark06 scores the users in that thread are getting, they are on par if not even higher then C2D? strange uh

[JumpingJack]

according to this post, by end user review Phenom is clock per clock good afteral

in supcom 2.64ghz + ~ 3GHz quad intel

that's not bad

LINK

also look at the 3dmark06 scores the users in that thread are getting, they are on par if not even higher then C2D? strange uh

You know what this reminds me of.... the P4 fanboys claiming P4 superiority when it won one or two multimedia benchmarks and PCMark.

[Epsilon84]

You know what this reminds me of.... the P4 fanboys claiming P4 superiority when it one one or two multimedia benchmarks and PCMark.

It's beardyman... don't mind him. He was the one that said Phenom 'PWNED' Core 2 in the face of indisputable proof.

He reminds me a lot of Sharikou...

[JumpingJack]

JJ,that explanation makes the most sense ,but it would imply(like you said) that L3 is a major bottleneck for K10,which is unlikely.It can make a big difference if NB is ran at the same as core speeds,but we need more tests to verify this.K10 looks and performs quite well,but looking at AMD section and some problems people have with early boards/BIOSes,it seems as the whole platform is still a diamond in a rough(problems with changing basic stuff in BIOS for instance).

Yeah, but I am looking for any reason why a 'logic' bug would only manifest itself at higher clock speeds.... that just does not make much sense to me. So far the two explanations I have that do make any sense is the NB clocking or a ridiculous hot spot which creates soft errors and results in a problem.

[BeardyMan]

It's beardyman... don't mind him. He was the one that said Phenom 'PWNED' Core 2 in the face of indisputable proof.

He reminds me a lot of Sharikou...

the funny thing is beardyman does own you with your own toys
Unlike you i'm neutral and i can admitt a defeat one brand by another.
Link

[JumpingJack]

Excellent link Primus_Sucks!

Hmm, so Phenom X3 @ 2.3GHz ~= E6750/E6850 in multithreaded performance.

Hopefully AMD can get clockspeeds much higher with B3 otherwise Phenom X3 will be stuck battling it out with the higher clocked Intel C2Ds.

Yeah, this is a good data set... they are showing ganged and unganged mode results too.... nothing much changes, except memory intensive stuff.. (Winrar shows some sensitivity).

It is likely it is a problem of heating at that specific location on the chip that would prevent the rise in frequency.

They are thinking a 'hot spot' to... if this is the case then a 'fix' will not improve performance at all.

[Syn.]

the funny thing is beardyman does own you with your own toys
Unlike you i'm neutral and i can admitt a defeat one brand by another.

why are you comparing Duel-Core against Quad-Core?

If you want my opinion on this i would say Phenom 9500 is like 2900XT, great idea just not made right. Which is the reason why i am waiting for the 3870 version of the Phenom which if history repeats it self will be great value for money.